Re: Fork in the road far back
"Meltdown was trivially addressed, and new os kernel models will improve performance closer to ignoring it."
The Meltdown mitigation is trivial only in concept. There is a massive performance difference between mapping kernel memory into each and every process's address space(SOP until now) and isolating it to its own, separate address space. This is expensive at the hardware level and always has been; there's no way for an OS to compensate.
As an analogy, let's replace syscalls with changing the volume on your TV. The modern standard for changing volume is with a remote; the Meltdown mitigation is roughly the equivalent of getting up each time you needed to change it. Needless to say, it takes a lot longer for the volume to be changed(syscalls to be completed) than before.
And before you say "performance, uh, finds a way", microkernels have been suffering from just this problem since they were conceived; if there was a solution, decades of research would've found it by now.